14 November 2026 · 4 min read
Anthropic Launches Self-Hosted Claude Managed Agents in Private Beta
Anthropic opened a private beta for self-hosted Claude Managed Agents on November 12, 2026, letting enterprises run the full agent service, sandboxes, and session history in their own infrastructure.
Anthropic released a self-hosted version of Claude Managed Agents on 12 November 2026, opening a private beta that lets customers run the agent harness, sandboxes, credentials and session history in their own cloud account or on-premises data centre [2][3]. Access is by request through an Anthropic account team; pricing has not been disclosed.
What changed
Until now, customers could self-host only the sandbox where an agent’s code executes. Anthropic still ran the agent loop, stored skills and memory stores, and handled credentials [3][4]. The new option moves the entire control plane into the customer’s environment. The customer now operates the Managed Agents server (cma-server), its PostgreSQL database, and the workers that run shell and file tools inside sandboxes. Session history lives in that database, and the server exposes the same Managed Agents API as the hosted service, bar a few documented differences [3].
How it works
The cma-server runs the agent loop, stores agents and their histories, and calls the configured inference provider. Separate workers execute tool code inside customer sandboxes; the server itself never runs those tools [3]. Developers build agents with the same core APIs, and the Claude SDK and ant CLI can connect to the customer’s server [3].
Model requests go to the Claude API by default. Customers can instead select Amazon Bedrock, Claude Models on Google Cloud (Google managed), Microsoft Foundry, or a model gateway they operate themselves. Agents continue to name models by their Claude API IDs, and the server translates for the chosen provider [3]. Each model call sends the full conversation – prompts, tool inputs and results – to that provider, so the LLM traffic still leaves the customer’s network unless routed through a self-managed gateway [2][3].
Credentials and security
Anthropic designed Managed Agents around four security principles that apply regardless of who hosts the service [2]. The harness that calls Claude is separated from the sandbox where code runs. Credentials stay outside the sandbox; for MCP tools and supported API calls, a proxy outside the sandbox attaches the credential to the request. Self-hosted customers can keep those credentials in their own secrets manager [2].
Each environment can be locked to an allowlist of permitted hosts. Requests to any other host are blocked. Customers who host the sandbox enforce those limits in their own firewall, making the allowlist their configuration to maintain [2].
Managed Agents logs every tool call and its result. The log exports over OpenTelemetry, so security teams can pipe agent activity into their existing monitoring tools [2].
Migration and operational responsibilities
Nothing is copied from Anthropic’s hosted service. Customers must recreate agents, skills, vault credentials and memory stores on their own server [3]. Organisations already running self-hosted sandboxes can connect the same worker software to the new server, although workers triggered by webhooks will not operate because the self-hosted server sends none [3].
Customers take on responsibilities that Anthropic holds for the hosted service: securing the server, hosting memory stores, detecting unusual usage and installing each release themselves [3]. Each deployment serves one tenant at one trust level. Separate tenants or workloads with different trust levels require separate deployments with their own databases, server API keys and workers [3].
Feature gaps
Several hosted-service features are unavailable in the self-hosted version: scheduled deployments, dreams and webhooks [3]. The web search and web fetch tools remain off until explicitly enabled, and their availability depends on the inference provider. Web search works with the Claude API, Google Cloud (Google managed) and Microsoft Foundry; web fetch works with the Claude API and Microsoft Foundry [3].
Session retention is customer-managed. Sessions stay in the database until deleted; deletion does not remove copies in backups, exported logs and traces, memory stores or, on Kubernetes, the session’s workspace volume [3].
What it means for developers
For teams building unattended agents that fix builds, test releases, triage alerts or work through bug backlogs, self-hosting puts the entire execution and state management inside their own perimeter. Credentials can stay in their own vault, and every step is auditable through OpenTelemetry. The trade-off is operational: the team must now run and patch the server, database and workers, configure allowlists correctly, and handle tenant isolation themselves.
The model itself still runs in the cloud, so prompts and completions leave the network unless a zero data retention agreement is in place and the inference provider is chosen accordingly [2][4]. For organisations that need full data sovereignty, that remains a boundary to watch.
Anthropic has not announced general availability or pricing. The private beta is the first step toward letting enterprises run Claude agents entirely on their own infrastructure, with the control and compliance burden that entails.
Sources
- Self-hosted Claude Managed Agents keeps agent credentials in your environment - Help Net Security
- Anthropic Lets Customers Self-Host Claude Managed Agents
- Anthropic's Gross Margin Battle Begins, Claude's Enterprise Penetration Becomes Key Bargaining Chip — BigGo Finance
- anthropic
- claude
- managed agents
- self-hosted
- ai agents
Researched and drafted with AI from the sources above.
Learn to build with AI
Hands-on tutorials on Cursor, Claude Code, AI agents and more.